Understanding interactive dynamic situations

A system is presented to interpret dynamic situations involving several moving objects in a known environment. It is part of the ESPRIT VIEWS project and processes data coming from sequences of video frames. These numerical data are interpreted to identify, in real-time, situations involving complex interactions between moving objects in the scene. The principle is to compile pre-defined behaviors for individual objects and then for groups of objects from elementary events. The method used is based on temporal networks and dynamic grouping. The aim of the method was to segment the recognition process into different layers according to the level of complexity of the items recognized. The applications studied concern the detection of incidents on an urban roundabout.<>